Financial Analytics: Lehman Brothers

Lehman Brothers is steadily rolling out new analytical models to support its in-house risk management systems for its exotic fixed income derivation business. The new models are part of a six-year risk project at the investment bank, which has been building ocmplex computational applications using the Linda and Paradise programming environents for parallel processing.

Paradise supports about 20 dual processors in Lehman Brothers' New York office, which run on Sun Microsystems' Solaris flavor of UNIX, says Thanos Mitsolides, Vice President of Fixed Income and Derivatives Technology at Lehman Brothers. Server clusters are also located in Leman's London and Tokyo trading offices. These serers create a computational pool for trade pricing and risk analysis of the fixed income derivatives products, which include American-style swaptions and index amortized swaps, he says. Trade data from the global clusters is harnessed for overnight batch risk analysis. This pool supports as many as 200 traders and analysts worldwise, he adds.

The Paradise-based computational pool produces and distributes overnight reports on as many as half a million deals. Lehman values trades on a deal by deal basis, rather than modelling entire portfolios, says Mitsolides. The investment bank produces the pricing reports in less time by using several cheaper Sun Unix servers, rather than downloading entire books into a large and costly Sun server by using parallel processing to value trade-by-trade, he says. "To make calculations on a good sized book would take about two to three weeks" to analyze well on one server," says Mitsolides.

Mitsolides says he evaluated higher-powered systems for the complex exotics pricing such as Cray supercomputers and languages such as Corba, but chose to work with Linda and Paradise because of its flexibility and its independence from hardware solutions.

The flexibility of Linda and Paradise, and the quickness of the trade valuations gives Lehman space to accommofate trader emergencies, says Mitsolides. Emergencies can be when a trader would like to model a portfolio in anticipation of a sudden market move or when a trader requests a different greek scenario for a trade, he says.

Linda and Paradise can also be easily programmed to access additional resources if needed. Such as, if a New York computational pool requires more processing power to handle a complex analysis on a high volume of data, a server from Lehman's London pool can be brought in to support the extra load, says Mitsolides.

The front end for Lehman's exotic fixed income derivatied traders is an in-house developed client server system built using Visual Basic, says Mitsolides. The trader front end allows for trade type definitions, hedge creation and Delta, Gamma, and some Vega analysis, he says.

Mitsolides was introduced to the Linda language when he was a graduate student at the Courant Institute of Mathematical and Computer Sciences at New York University.

Lehman piloted the Kdb database management system from Palo Alto-based Kx Systems (D&RT, May 11, 1998). Lehman had been evaluating whether to use the Kdb database to pool static and historical information on fixed income products.